AI companies need people to write, check and rank the answers their chatbots give. That work is called AI training, data annotation or RLHF (reinforcement learning from human feedback), and it's one of the fastest-growing kinds of remote work. Some of it pays well. Much of it is unsteady. This guide covers what the main platforms pay, who they accept, and how to tell a real platform from a scam.
The short version: general AI training pays roughly $15 to $30 an hour on most platforms, and specialists (coding, maths, medicine, law, finance) earn $40 to $150. Work comes in waves, there's no contract for hours, and many of the best-paid projects are limited to the US and a few other countries. Treat it as flexible extra income, not a replacement for a steady job.
What the work actually is
Most tasks are one of these:
- Rating answers. You see two chatbot replies and pick the better one, then explain why.
- Writing prompts and ideal answers. You write a question, then the answer a perfect assistant would give.
- Fact-checking. You check a reply against sources and flag anything wrong.
- Labelling. You tag images, audio or text (for example, "this sentence is angry").
- Expert review. Coders check code, nurses check medical advice, accountants check tax answers.
Strong written English, attention to detail and following long instructions exactly matter more than any qualification. If you've done chat or email support, you already have the core skill: reading carefully and writing clearly.
What the main platforms pay
Rates below are what each platform advertises, checked in October 2026. Platforms change rates often and show you the rate before you start a project.
| Platform | Advertised pay | How you get in | Pays through |
|---|---|---|---|
| DataAnnotation | General $25–$50/hr; coding $40–$150+; STEM and professional $40–$125+ | One starter assessment, about an hour (you only get one try) | PayPal |
| Outlier | Generalist up to about $15/hr; experts up to $90–$150/hr | ID check, CV and LinkedIn, skill tests; needs at least an associate degree | PayPal, Airtm or ACH, weekly |
| Mindrift | About $15–$30/hr entry; $60–$100+ for specialists | Expert in one of 90+ fields, ID check | Payoneer or PayPal, twice a month |
| Alignerr | $30–$120/hr depending on field | CV, AI video interview, tests, ID check | Deel, PayPal or Stripe, weekly |
| Prolific | Set per study; at least $8/hr, $12/hr recommended | Waitlist, ID check | PayPal ($6 minimum cash-out) |
| Mercor, Handshake AI, Snorkel | $20–$200/hr for specialists | Interviews and background checks | Mostly US-only |
Who actually gets accepted
This is where most guides stop being honest. Three things decide whether you get work:
- 1.Where you live. Many top-paying projects are US-only (Handshake AI, Snorkel and most Mercor roles say so). Outlier and DataAnnotation accept people in many more countries, but some projects are limited by region, and the rate can depend on your location.
- 2.Your background. Generalist queues are crowded. A degree or real experience in coding, maths, science, medicine, law or finance moves you to far better-paid and steadier work.
- 3.Your first test. DataAnnotation's starter assessment can only be taken once. Prepare as if it were a job interview: read every instruction, write full reasons, and don't rush.
If you speak a second language well, look for multilingual projects. DataAnnotation advertises $25 to $40 an hour for them, and fewer people can do them.
The honest downsides
- No guaranteed hours. Queues go empty for days or weeks, often with no explanation.
- Sudden removals. Workers on several platforms report being removed from projects or deactivated without warning.
- Unpaid time. Onboarding, tests and rejected tasks often aren't paid.
- You're a contractor. No paid leave, no sick pay, and you handle your own tax. Read contractor vs employee.
- Legal questions. Several platforms face lawsuits over whether their workers should be employees. Outlier's owner settled one in California for $12.5 million.
That's why the best way to use AI training is alongside a steady remote job, not instead of one.
How to spot a fake AI training job
Scammers love this trend. The US Federal Trade Commission says reported job-scam losses rose from $90 million in 2020 to over $500 million in 2024, and "task scams" were the biggest single type. Walk away if:
- Someone messages you first on WhatsApp, Telegram or text about "AI rating" or "app optimisation" work.
- You're paid a few dollars at first, then asked to deposit money or crypto to "unlock" tasks or withdraw earnings.
- The "job" is liking, rating or "boosting" products.
- The site isn't the platform's official domain.
Real platforms never ask you to pay. More signs in how to spot remote job scams.
How to get started this week
- 1.Pick two platforms that accept your country and sign up only on their official sites.
- 2.Practise your writing. Most tests reward clear, well-reasoned explanations. Our free English test grades a written reply and shows your level.
- 3.Take the starter test when you're fresh, with a few quiet hours free.
- 4.Get paid cheaply. Most platforms pay through PayPal, Payoneer or Airtm; compare what you keep with the pay calculator.
- 5.Build a steady base. Pair AI training with a remote job that has set hours and set pay.
A steadier option: remote support jobs
If you can write clear, careful answers for an AI, you can do the same for real customers, with a contract, regular shifts and pay you can count on. Every job on Remote shows its hourly pay in US dollars and which countries it's open to, so you only apply where you can be hired.
Create a free account to see roles open to your country, or browse remote jobs open worldwide.





